Subject: Re: SCC and (deterministic) counterfactuals

Following the article and commentary on SCC, which I have read this morning.

Unfortunately, the word "determinism" never shows up in the article and commentary. Of course, that's exactly where the SCC comes from.

Interestingly, every physicist knows that Laplace's model of a deterministic universe was replaced by evidence of an indeterministic micro-cosmos (as found in quantum mechanics.) Epi methodologists seem to have not caught up yet. They still talk about a sufficient cause behind every case of lung cancer, and derive volumes of methodological implications from a possibly erroneous foundation.

More important, why does everyone embrace the helpful consequences of SCC and ignore the embarrassing derivations? You surely know much of what I wrote in the attachment (selected sections from book chapters). Years ago I spent several hours with George Maldonado ("Estimating causal effects") showing him where SCC and deterministic counterfactuals take us. He had no good answers. Maybe you can help me see a more optimistic way out of the deterministic conundrums?

By the way, there are ways to reconcile the principles of DAG with indetereministic causation.

Thanks for your time,

Eyal Shahar, MD, MPH

Professor

Division of Epidemiology and Biostatistics

Mel and Enid Zuckerman College of Public Health

The University of Arizona

Hi Dr Shahar,

Thank you for your thoughtful email. In it and the attachment, you raise some interesting questions concerning determinism. Because of a long-standing interest in physics, I too had thought about some of these issues - like the Heisenberg Uncertainty Principle, although I think not in as much depth as you.

I have attached some brief - and incomplete - thoughts on some of these issues.

If you find them useful and stimulating, perhaps we can continue to correspond and pursue these issues.

Sincerely,

XXXX

Hi Dr.XXX,

Thank you for your email and your attached writing, which I found interesting. I will be happy to continue to pursue the issues by email.

Before trying to tackle the task of coherent causal inquiry in an indeterministic universe (about which I have some thoughts), perhaps we can clarify first the devastating implications of determinism. That was never bluntly stated by reputable methodologists. The excerpts I sent you tried to state just that. In brief, the key derivations are as follows:

1. There are no universal, quantitative laws of causation in a deterministic universe. There are no "universal causal parameters".

2. A causal parameter is defined on a "target population" (a real list of names!). Not to be confused with vague notions of "super-population" or "near-infinite population". The latter are not "target populations".

3. Diverging results of studies are the natural expectation from a deterministic model of causation—-not the exception—-since there is no reason to expect similar proportions of causative, preventive, (and doomed) in different studies. In short, there is no obvious reason to call conflicting results of different studies "conflicting results." Different studies estimate different target-specific parameters, and there is no a priori reason why causal parameters of different "target populations" should be identical or even similar.

4. Consistency of empirical results (which is not uncommon) is an unexplained miracle of nature.

5. Effects should be measured on additive scales. Our multiplicative models (logistic, Cox) are inferior to additive models. Ratios are the wrong measures of effect (and so is interaction on multiplicative scales).

6. Effect modification on the additive scale is not a surprise. It is the expectation. (In fact, we should not use the term "effect modification" at all--the correct word should be "interaction"). We should be surprised when we DON'T find interaction. Again, its absence is an unexplained miracle of nature.

All of these derivations follow from the fundamental idea of "causative", "preventive", "doomed", and "immune", which is derived from the SCCM (i.e., determinism). They are derived by trivial math and logic, and I am sure that people like Greenland, Rothman, Poole, Maldonado, and others clearly know them. What I haven't found in their writing so far is head-to-head confrontation with the epistemological consequences--with what I call "the epistemological poverty of determinism". In other words: determinism reduces causal knowledge to "names of causes" and names of "interacting causes" plus an endless list of target-specific parameter estimates.

Maldonado, for example, agreed verbally with me that consistent results are not expected under the "4-types" model. But he never wrote it. Greenland injected the term "target population" into Modern Epidemiology (and articles), but did not explain clearly that it is a finite list of names. Rothman (I suspect) can't stand the idea of "target population", but realized that it must follow the SCCM (his pies of component causes).

Looking forward to hearing more (when you get a chance).

Thanks again for your interest.

Best,

-Eyal

Eyal Shahar, MD, MPH

Professor

Division of Epidemiology and Biostatistics Mel and Enid Zuckerman College of Public Health The University of Arizona